Equilibrated Averaging Residual Method

• EARM is a unified flux recovery framework for finite element methods that employs a two-step averaging–residual correction to achieve robust, locally equilibrated flux reconstruction.
• It is explicitly constructed to work with conforming, nonconforming, and discontinuous Galerkin discretizations in both two and three dimensions.
• The method ensures local conservation and optimal a posteriori error estimators while maintaining computational efficiency and robustness against large coefficient jumps.

The Equilibrated Averaging Residual Method (EARM) is a unified flux recovery framework for finite element solutions of elliptic interface problems. EARM enables explicit, locally conservative, and robust flux reconstruction for conforming, nonconforming, and discontinuous Galerkin (DG) discretizations of arbitrary polynomial order in two and three dimensions. Its defining feature is a two-step averaging–residual correction, producing equilibrated fluxes suitable for sharp and efficient a posteriori error estimation. Methodological flexibility, explicitness, and robustness under large coefficient jumps are central characteristics, as established in recent literature (He, 5 Mar 2025, He, 4 Jan 2026, He et al., 2020).

1. Motivation and Context

A posteriori error estimation strategies for finite element methods (FEM) often require the construction of discrete fluxes $\sigma_h \in H(\div;\Omega)$ satisfying two key conditions on each mesh element $K$:

• Local conservation: $\nabla \cdot \sigma_h|_K = f_K$,
• Proximity to the raw FE flux: $\tau_h = -A \nabla u_h$, so that $\|A^{-1/2}(\sigma_h + A \nabla u_h)\|$ controls the energy-norm error.

Classical residual and patch-based flux recovery methodologies impose element-specific or ad hoc constructions tailored to particular FE spaces (CG—conforming Galerkin, NC—nonconforming, DG—discontinuous Galerkin), creating barriers to generality and explicit computation. EARM was introduced to overcome these obstacles, providing a robust, fully general procedure applicable to all standard FE discretizations (He, 5 Mar 2025, He, 4 Jan 2026).

2. Mathematical Formulation

Let $\Omega \subset \mathbb{R}^d$ be partitioned by mesh $\mathcal{T}_h$, with $A(x)$ symmetric, positive-definite, and piecewise constant. For any given solution $u_h$, EARM proceeds in two main stages:

(A) Weighted Averaging Step: Construct a flux $\tilde{\sigma}_h \in RT(\mathcal{T}_h,s)$ (Raviart–Thomas space of order $s$), defined via facet normal moments and, when $s \geq 1$, by matching low-order cell moments: $$\int_{F} (\tilde{\sigma}_h \cdot n_F) \, \phi \, ds = -\int_{F} \{A \nabla u_h \cdot n\}_w^F \phi \, ds,$$ for all basis functions $\phi \in \mathbb{P}_s(F)$, with similar constructions on Neumann boundary faces.

(B) Residual Correction Step: Evaluate the elementwise residual functional: $$R_{s,K}(v) := (f - \nabla \cdot \tilde{\sigma}_h, v)_K, \qquad v \in DG(\mathcal{T}_h,s),$$ Then define a correction flux $\sigma_h^\Delta \in RT(\mathcal{T}_h,s)$ via conservation constraints: $$(\nabla \cdot \sigma_h^\Delta, v)_K = R_{s,K}(v), \qquad \forall v \in \mathbb{P}_s(K),\; \forall K,$$ giving the recovered flux $$\hat{\sigma}_h = \tilde{\sigma}_h + \sigma_h^\Delta$$. This procedure ensures $\nabla \cdot \hat{\sigma}_h = f_h$ elementwise, with $\hat{\sigma}_h$ conforming in $H(\div)$ (He, 5 Mar 2025, He, 4 Jan 2026).

3. Euler–Lagrange Framework and Solution Procedure

EARM’s correction stage admits a variational description: $$\begin{aligned} & (A^{-1} \sigma_h^\Delta, \tau)_\Omega + (\lambda_h, \nabla \cdot \tau)_\Omega = 0, \quad \forall \tau \in RT(\mathcal{T}_h, s), \ & (\nabla \cdot \sigma_h^\Delta, v)_\Omega = R_s(v), \quad \forall v \in DG(\mathcal{T}_h, s), \end{aligned}$$ where $\lambda_h$ acts as a Lagrange multiplier. In many cases, especially DG and nonconforming settings, flux corrections can be constructed explicitly through moment-matching without solving a global system. For conforming elements, ON-EARM (Orthogonal Null-space–Eliminated EARM) restricts the correction to the $L^2$-orthogonal complement of the divergence-free null space, ensuring uniqueness (He, 4 Jan 2026).

4. Local Conservation, Explicitness, and Algorithmic Aspects

By design, EARM achieves exact elementwise equilibrium for the recovered flux: $$\int_{\partial K} \hat{\sigma}_h \cdot n_K \, ds = \int_K f, \quad \forall K.$$

Algorithm Variants by Discretization:

Element Type	Averaging Step	Correction Step	Result
CG($k$)	$\tilde{\sigma}\in RT(k-1)$	Global/patchwise symmetric system; ON-EARM	Unique equilibrated flux
NC($k$) (odd)	$\tilde{\sigma}$ via edge moments	Explicit flux jumps, closed-form formulas	Explicit equilibrated flux
DG($k$)	$\tilde{\sigma}$ via facet averaging	Explicit elementwise RT($k$) problems	Explicit equilibrated flux

Explicit two-dimensional reconstructions are available for all cases, with small local systems required in three-dimensional CG settings (He, 5 Mar 2025, He, 4 Jan 2026).

5. A Posteriori Error Estimation and Theoretical Properties

EARM facilitates robust, locally efficient, and globally reliable a posteriori estimators with Prager–Synge-type identities as the foundation. For any equilibrated flux $$\hat{\sigma}_h \in RT_f:\; \nabla \cdot \hat{\sigma}_h = f_h$$, local indicators are defined: $$\eta_{\sigma,K} := \|A^{-1/2}(\hat{\sigma}_h + A \nabla u_h)\|_{L^2(K)},$$ with the global estimator $\eta_\sigma = (\sum_K \eta_{\sigma,K}^2)^{1/2}$. Reliability is ensured by the identity: $$\|A^{1/2}(\nabla u - \nabla u_h)\|^2_\Omega \leq \inf_{\tau \in RT_f}\|A^{-1/2}(\tau + A \nabla u_h)\|^2 + \inf_{v \in H^1_D}\|A^{1/2}(\nabla v - \nabla u_h)\|^2,$$ and taking $\tau = \hat{\sigma}_h$ recovers the estimator. Efficiency constants are independent of the jump in $A$ under mild "quasi-monotonicity" conditions (He, 5 Mar 2025, He, 4 Jan 2026, He et al., 2020).

6. Practical Implementation and Computational Cost

EARM is distinguished by fully explicit and highly parallelizable algorithms for DG and nonconforming discretizations. For DG($k$) and odd-order NC($k$), all local problems are solved in small fixed-size polynomial subspaces; for CG elements, ON-EARM yields a global symmetric positive-definite system on facet unknowns, considerably smaller than the original FE system. Precomputation of local matrices and matrix factorizations further accelerates patchwise procedures in 2D (He, 5 Mar 2025, He, 4 Jan 2026).

7. Numerical Validation, Generalization, and Outlook

Numerical results in benchmark problems (e.g., Kellogg’s interface, L-shaped domains, Fichera corner in 3D) confirm that EARM-based estimators achieve optimal convergence rates and tightly track the true error, with effectivity indices typically ranging from $1.1$ to $6.6$ depending on order and mesh refinement. The method is robust against large jumps in the diffusion coefficient $A$. The framework generalizes across mesh dimension, element order, and discretization type, encompassing previously proposed explicit flux recovery and patch methods as special cases (He, 5 Mar 2025, He, 4 Jan 2026).

This suggests that EARM provides a canonical formulation for locally equilibrated a posteriori error estimation, underlining its utility for adaptive mesh refinement and reliability in multi-physics simulations with heterogeneous coefficients.